Language Edition

Language Edition turns one admitted timed-media work into an attributable language edition. It owns the language-edition project, source-aligned words, target-language judgment, casting and performance direction, dialogue timing, captions, and the acceptance of a constrained language-track replacement. Speech engines, generic translation, Persona identity, music, visual authorship, editorial recutting, and publication remain with their own offices.

The target language may equal the source language for replacement dialogue, narration, repair, or accessibility. Translation is therefore an optional semantic act inside the Pattern, not the definition of dubbing and not a reason to create an ownerless Translation Composition.

Contract

Language Edition qualifies as a Composition because the project survives any one engine or Product. Segment identity, source/target correspondence, glossary choices, casting, pronunciation, timing, disclosure, review findings, accepted performance revisions, unresolved regions, and recovery all remain durable application truth. A Product may package Language Edition by itself, but need not reuse its technical name as the market name. Reusing Echo, Prism, Scout, or a translation implementation does not turn those Extension Domains into the application.

One bounded language-edition road

optional upstream acquisition (outside Language Edition)
→ source locator
→ Scout Fetch/Download + separate Artifact Admission
→ immutable source master + exact timebase handoff
Language Edition Invocation begins
→ bounded demux/probe
→ Echo Ear transcript, diarization, and timing observations
→ source-aligned DialogueTranscriptRevision@1
→ optional attributed Translation Spell + human or admitted review
→ LanguageEditionScriptRevision@1
→ casting from exact performer/voice eligibility references
→ human takes or Echo Voice synthesis
→ DialoguePerformanceRevision@1
→ dialogue edit, alignment, captions, and DialogueConformMap@1
→ intelligibility, linguistic, timing, rights, and disclosure acceptance
→ TimedLanguageAssetBundle@1
→ optional constrained language-track mux against the locked master
→ probe and back-check → LanguageVersionCandidate@1 or an exact finding

Every arrow is separately admitted. A URL is not media custody, STT is not a canonical script, detection is not permission to translate, translated prose is not an accepted performance, TTS generation is not playback, and a successful mux is not publication. A provider failure never authorizes another language, voice, model, road, or paid service.

The Language Edition mux may replace or add only the declared dialogue, narration, caption, subtitle, and language-metadata tracks against an exact locked source master. It records preserved and replaced stream digests, timebases, codecs, channel maps, offsets, loudness and intelligibility probes, and the resulting digest. A new cut, reordered scene, changed claim, newly authored picture sound, or different music is editorial work for Broadcast, not a convenient mux option.

TimedLanguageAssetBundle@1 is the accepted downstream handoff: exact source and edition-script revisions, performance and dialogue-track artifacts, captions, conform map, rights/disclosure posture, findings, lineage, and admission criteria. It is not a recut or final picture container. LanguageVersionCandidate@1 is an optional restricted derivative made by muxing that exact bundle against one locked master while preserving every non-language stream. Broadcast or another consumer may admit the bundle into its own assembly without treating the candidate as canonical.

Translation is a Spell, localization is the judgment

Spellweaver can validate and place a versioned Translation Spell, but it does not own linguistic truth. The Spell's authority-qualified publisher owns its generic transformation semantics and revision; the placement pins original text, source and target languages, glossary and protected spans, implementation revision, detection and fallback policy, review requirements, and declared loss. Language Edition retains both source and derivative and decides whether the result is fit for this speaker, scene, audience, duration, and edition.

Another Composition may reuse the same exact Translation contract while retaining its own records, criteria, and finish judgment. Website and Altar chrome localization remains a Frontend catalogue concern; documents, product copy, and other untimed writing remain with their content owner. Language Edition owns captions, subtitles, scripts, and on-screen-language cue sheets only when they are part of its timed language edition. The FOSS reference Altar can therefore remain English-only without making multilingual media impossible or baking English grammar into canonical identities.

Voices belong to no shortcut

Echo owns speech capture, transcription, synthesis, delivery, and their chronology. An Animator profile names the exact engine, model or acoustic voice, languages, formats, licence, and measured limits. Performer consent and source rights remain exact inputs; neither an attractive result nor a profile name grants them.

Language Edition owns only the casting of an eligible reference to a role in this edition, its direction, pronunciation, timing, disclosure, and performance acceptance. Avatar may reference an eligible voice in a Lich presentation profile or supply that exact eligibility as an external precondition when the Lich is being dubbed. Language Edition cannot revise the Persona, broaden the voice use, mint identity from resemblance, or make one approved line authorize arbitrary later words. Riffmaw retains sung performance and music production rather than general speech.

Neighbours keep their truth

An already-settled reference crosses each seam without a Suite. A Suite is needed only when one promised result must actively admit, await, cancel, retry, recover, or jointly settle new Language Edition, Voidlight, Riffmaw, Avatar, or Broadcast Invocations.

Failure, restart, and the first proof

Each segment has a stable identity against the exact source digest and timebase. Transcript, translation, performance, alignment, dialogue mix, restricted mux, and review attempts retain their own revisions and terminal dispositions. Unsupported language, ambiguous speaker, missing rights, excessive timing loss, failed intelligibility, stale source, exhausted budget, or indeterminate provider/write effect returns a finding, refusal, partial edition, or unknown; none is silently repaired by substitution. Restart resumes only from reconciled immutable artifacts and settled attempts.

The smallest proof is local, synthetic, and network-disabled: one short two-speaker video with a known timebase and separate picture, dialogue, music-and-effects, and caption fixtures. Produce one same-language replacement and one translated edition through fake Ear, Translation, and Voice implementations; preserve the locked picture and non-language stream digests; prove segment lineage, glossary protection, one human correction, timing pressure, caption alignment, partial speaker refusal, crash recovery, deterministic mux, export, and deletion. It proves the Composition contract—not translation quality, a real voice model, performer rights, publication, or delivery.

Related: Workflow · Audio · Vision · Composition Portfolio